Vasara

Soft-delete a project and enqueue cleanup. Owners only. Destructive write. The Projects App calls this after the user acts on the project list. If the user wants to create, rename, or delete a project in the UI, you call open-projects. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

delete-project

Delete a project document. Pass projectDocumentId as the opaque token returned by a project document list or search. The Projects App calls this after the user acts on the documents page. If the user wants to upload, rename, or delete files in the UI, you call open-project-documents. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

delete-project-document

You find later court rulings that mention a case number (token/phrase match on chunk text — not semantic search). You use this as a citator to see how a judgment is cited in other domestic rulings. You do not apply the searchCaseLaw 3/6-year retry windows to this tool. Optional dateFrom/dateTo narrow the citing-document window. If those dates are omitted and the source case is found, dateFrom defaults to that case's date. After you use this tool, you do not re-run searchCaseLaw with the same case number as the query. You use getCourtCaseByCaseNumber only when you need the full text of the source case itself. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

find-case-mentions

Returns the authenticated user's PostHog identity for MCP App analytics. Host Apps call this; do not use it in conversation.

get-mcp-app-analytics-identity

You fetch the full text of a court judgment when the user gives an exact case number (document title match). You use this tool to verify the validity, relevance, or current status of specific court cases mentioned in a document or prompt. This call is expensive. You use it only when full text is required. If the user names the court, you pass it as `courtName`, because several common courts use the same case number. You cite the judgment with its `citationToken` as `[<documentTitle>](/?chunk=<citationToken>)`; you never cite `documentId` as a chunk. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

get-court-case-by-case-number

Return insight grid metadata: columns, rows (document names), and cell status counts. Does not include full cell text — call queryInsightGridCells for values. Pass gridId (opaque token from list-project-insight-grids) on MCP; insight-grid assistants may omit it when a grid is already pinned. When sourceLink is present (MCP), cite the grid with that URL. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-insight-grid-snapshot

Load the full insight-grid workspace snapshot (columns, rows, cells, documents) for the Projects App. Poll this while a run is in flight. Text-only — do not confuse with get-insight-grid-snapshot (metadata only) or calculate-insight-grid (local aggregations). If the user wants to work in one insight grid in the UI, you call open-project-insight-grid. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-insight-grid-workspace

Load a legal corpus source preview for the Projects App source-preview panel. Pass chunkUuid and/or sourceId. App-only — not for agent research; use legal search tools instead. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-legal-source-preview

Load chrome for one project (name, role, canEdit, document and insight-grid counts). The Projects App uses this for in-app navigation. You do not attach UI. If the user wants the Projects App UI, you call the matching open-* tool. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-project

Fetch the full converted text content of a project document by projectDocumentId (opaque token from list/search — not a project id). No projectId arg required. When sourceLink is present (MCP), prefer that URL for citations. Returns deleted=true when the document has been removed. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-project-document-content

Load preview data for a project document. Binary previews return a short-lived bytesUrl; text and HTML previews return content. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-project-document-preview

List all documents in a project. Pass projectId (opaque token from list-projects) on MCP; in-app assistants may omit it when a project is already pinned. Returns projectDocumentId tokens for get-project-document-content — do not use a project id there. When sourceLink is present (MCP), prefer that URL for citations. If the user wants to browse, upload, or manage project files in the UI, you call open-project-documents. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

get-project-documents-list

Vasara legal research protocol. ## Parallel tool calls You issue all independent searches in one parallel batch — same or different tools in a single step. You do not call a tool, wait for the result, then call another when you already know both searches are needed and their inputs do not depend on each other. ## Rules of the active legal jurisdiction - The user's Vasara account sets the active legal jurisdiction. The rules below apply to every legal jurisdiction. - Before your first legal search in a conversation, you call this tool once. Its reply gives the jurisdiction lock, the query language, the courts, the tools, and the citation examples of the active legal jurisdiction. - Every legal search result starts with `jurisdictionLock`. If a rule here and `jurisdictionLock` differ, you follow `jurisdictionLock`. ## Source Strategy 1. **MASTER SEARCH PROTOCOL:** You follow these steps for every legal question. **STEP 1 — First search wave:** You always call `search-laws` first. Exception: If you only verify or extract specific court cases by case number, you call `search-case-law`, `get-court-case-by-case-number`, or `find-case-mentions`. You do not need `search-laws` first in that case. If the user names specific document titles, you also call `search-chunks-from-document-by-title` in parallel. If the question names a code and an article number (e.g. {code} {article}), you also call `search-chunks-from-document-by-title` on the official code title with a number-only article query in that same first wave. If the question involves disputes, court decisions, complex practical scenarios, or rights and obligations in practice, you also call `search-case-law` in parallel for domestic judicial interpretation. **STEP 2 — Evaluate:** After each search wave, you ask whether you found an explicit rule that answers the question. A definition without a permission or a prohibition is not enough. If you found only context, you keep searching. **STEP 3 — Tool switching:** If results are inadequate, you call a different tool before you call `search-laws` again. You alternate with `search-legal-acts`, `search-case-law`, `search-organization-documents`. **STEP 4 — Persistence:** You attempt up to **4 substantive search iterations** with varied keywords and tools before you conclude that sources are absent. **STEP 5 — No early quitting:** If you deliver a shallow interpretation after only one unsuccessful search, that is a failure. 2. If nothing relevant exists after exhaustive search (per above), you state that the corpus did not yield an answer. You do not invent or infer an answer from your general knowledge. You do not claim 'in practice it means...' (e.g. 'courts usually hold...') unless you cite specific cases that establish those practices. 3. **Case-law date windows:** For case-law search tools that take `dateFrom` / `dateTo`, you start with `dateFrom` about three years ago. If the result is empty or insufficient, you retry with a six-to-three-year window. If it is still empty, you make a final pass with an upper bound of six years ago. Do **not** apply these 3/6-year windows to `find-case-mentions`. 4. **vectorSearchQuery is a single string (never an array):** Every search tool requires `vectorSearchQuery` to be one non-empty query string in the query language of the active legal jurisdiction — never the user's UI or message language. If the handed question is in another language, you translate into the query language of the active legal jurisdiction first. You never pass an array (and never `[]`) for `vectorSearchQuery`. You do not call a search tool until that one query string is ready. 5. **Batched queries (preferred):** To run several alternative queries with identical filters in a single call, you keep one in `vectorSearchQuery` and put the rest (1–7 more) in the optional `additionalVectorSearchQueries` array. The runtime embeds and executes them in parallel, interleaves them, and dedupes them. Each distinct query adds up to 6 snippets (maximum 18 total). Each returned snippet includes `matchedQueries` listing which query string(s) found it. You use that list to judge which search angle produced the evidence. Multiple matches indicate consensus. If you have only one query, you omit `additionalVectorSearchQueries`. You never send it as an empty array. 6. If the same hypothesis needs different filters (e.g. different `court`, `dateFrom`, `documentId`), you issue parallel duplicate calls. You use `additionalVectorSearchQueries` only when filters are identical. 7. If early tries return zero hits, you widen keywords before you give up. 8. You retry a failed tool call once with unchanged args. After that you report incapacity. 9. **No duplicate searches:** The runtime hard-deduplicates tool calls within a single run. If you issue the same tool with the same input, or any call whose entire result was already returned by an earlier call in this run, the runtime returns a `DUPLICATE SEARCH SKIPPED` / `All matched chunks were already returned` message and zero new content. You treat that signal as a pivot cue. You switch the tool, the hypothesis, the query angle, the filters, or the date window. You do not re-issue the same search. 10. **Complete parameters:** You fill every mandatory field in the schemas. ## Source links in answers - You always link your sources in the final answer whenever a URL is available. You do not cite legal sources as plain text only if a link can be included. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. - **Citation labels:** You write the citation label. Keep the citation label at or under 70 characters. Use retrieved snippet fields `title`, `documentTitle`, `text`, `organizationName`, and `reference`. Do not use `documentType`. Do not use a citationToken as the citation label. - **Known codes** (official code abbreviations of the active legal jurisdiction): `{code} {article} {heading}`. Include the heading only if it appears in this act's `text`. If the number is in `text` but no heading title: `{code} {article}`. If no number is visible: `{official code title}`. Do not invent an article. - **Other statutes / legal acts:** `{title} {article} {heading}` when a live heading is in `text`. Otherwise `{title}` or `documentTitle`, plus `reference` when it helps. - **Domestic rulings:** `{court} {title}` where `title` or `documentTitle` is the case number. Example: `{court} {case number}`. You take the court from `organizationName` (the court codes of the active legal jurisdiction, or a short court name). If several orgs are joined, you use the court. You do not use a bare case number when the court is known. You do not use a generic "{court} judgment" when the case number is present. - **ECHR / CJEU:** `ECHR {applicant} v. {state}` / `CJEU C-{number}/{year}`. Trim long party strings. Use `title` or `documentTitle`. - **EU acts / commentaries / templates / MGK:** `{title}`. Add `reference` only when it disambiguates. Skip commentary URL-hash suffixes. - **Uploads / web / projects:** file name + clause; page `name`; project/grid `name`. - If multiple snippets support the same point, include a link for each distinct source where `sourceLink` (or an organization URL) is available. - When describing which redaction of a law you used, prefer snippet `editionValidFrom` over `documentDate`. `documentDate` is the original act date; `editionValidFrom` is when the retrieved consolidated edition entered force. - **Article numbers only from live-act snippet `text`:** Retrieved snippets have **no** separate article-number field. If you name a statute article or paragraph (e.g. "{code} {article}"), that number must appear in the retrieved live-act snippet `text` as an explicit heading such as the article heading form of the active legal jurisdiction (in that retrieved snippet or another retrieved snippet of the same act). You do not invent, recall, or transpose article numbers from pre-trained memory of older consolidations. - **Case-law article cites are historical until verified:** A judgment snippet often quotes article numbers from the redaction in force when the case was decided. You treat those numbers as historical coordinates inside that judgment. You do not treat them as current statute anchors. Before you state "under article X of the act" as live law, you retrieve the current act text. You use only an article number that appears in that live-act snippet `text`. You may still cite the case for its interpretive holding (e.g. a high-court holding + "in the current version — {article}" with the act chunk once the heading is visible). - **Mid-article chunks (no article heading in `text`):** If a useful retrieved snippet starts mid-provision (e.g. starts at "9. {provision text}…" with no article heading), you do not invent the article number from memory or from a judgment snippet. Required order: (1) you search the same document by title or neighboring queries until a retrieved snippet `text` shows the article heading for that provision, then you cite that live number; (2) if the heading still never appears, you cite the act and chunk **without** a specific article number (e.g. `[{act title}](/?chunk=…)` describing the rule in prose). You do not guess an article number or any other number. - **Fatal mismatch:** If you use a live-act `citationToken` and you label it with an article number that does not appear in that retrieved snippet `text` (or in another same-act retrieved snippet you also cite for the heading), that is a FATAL ERROR.

get-legal-research-instructions

List organization users for the project member picker (tokenized ids, name, email). Not for inviting people outside the org. The Projects App calls this after the user acts on the team page. If the user wants to add or remove members in the UI, you call open-project-team. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

list-organization-users

List all insight grids in a project with row and column counts. Pass projectId (opaque token from list-projects) on MCP; in-app assistants may omit it when a project is already pinned. Returned grid ids are opaque tokens for get-insight-grid-snapshot / query / calculate. When sourceLink is present (MCP), cite grids with that URL. If the user wants to create or manage insight grids in the UI, you call open-project-insight-grids. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

list-project-insight-grids

List members of a project (tokenized user ids, name, email, role). Text-only; the Projects App uses this for team-page navigation. If the user wants to view or manage the team in the UI, you call open-project-team. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

list-project-members

List accessible projects as text only (no picker UI). Use this when the user already named a specific project and you need ids/tokens to match that name. If the user wants to create a project, you call open-projects. If the user wants to browse the project list in the UI, you call open-projects. If you cannot determine which project they mean, call select-project-or-grid. You do not ask them to type a name. Project ids are opaque user-scoped tokens — pass them as projectId to open-* or list tools. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

list-projects

Open the Projects App on the documents page. On that page the user can browse files, upload files, preview a file, rename a file, delete a file, retry a failed ingestion, and download a file. If the user wants to browse, upload, or manage project files in the UI, you call this tool. Pass projectId from list-projects, or documentId from a project-document citation to open that file. Optional quote and page seed the file preview. Returned document ids are opaque tokens. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

open-project-documents

Open the Projects App on one insight grid. On that page the user can edit columns and rows, set cell values, run insights, clear insights, and preview files and legal sources. If the user wants to work in one insight grid in the UI, you call this tool. Pass gridId from list-project-insight-grids. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

open-project-insight-grid

Open the Projects App on the insight-grid list. On that page the user can browse grids, create a grid, rename a grid, and delete a grid. If the user wants to create or manage insight grids in the UI, you call this tool. Pass projectId from list-projects. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

open-project-insight-grids

Open the Projects App on the team page. On that page the user can view members, add organization users, and remove members. If the user wants to view or manage the team in the UI, you call this tool. Pass projectId from list-projects. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

open-project-team

Open the Projects App on the project list. On that page the user can browse projects, create a project with the New project control (name, members, and files), rename a project, and delete a project. If the user wants to create, browse, rename, or delete projects, you call this tool. If the user wants to create a project, you call this tool. You do not call `create-project` for that request. list-projects stays text-only when you only need to match a named project. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

open-projects

Query insight grid cell values with optional filters by column/row IDs or indices, status, or tag flag. Pass gridId from list-project-insight-grids on MCP; insight-grid assistants may omit it when pinned. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

query-insight-grid-cells

Remove document rows from an insight grid. Destructive write. The Projects App calls this after the user acts in one insight grid. If the user wants to edit columns, rows, cells, or run insights in the UI, you call open-project-insight-grid. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

remove-insight-grid-rows

Remove a project member. Cannot remove yourself. Destructive write. The Projects App calls this after the user acts on the team page. If the user wants to add or remove members in the UI, you call open-project-team. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

remove-project-member

Rename a project document. Pass projectDocumentId as the opaque token returned by a project document list or search. The Projects App calls this after the user acts on the documents page. If the user wants to upload, rename, or delete files in the UI, you call open-project-documents. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

rename-project-document

Retry ingestion for a project document whose status is error. Pass projectDocumentId as the opaque token returned by a project document list. The Projects App calls this after the user acts on the documents page. If the user wants to upload, rename, or delete files in the UI, you call open-project-documents. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

retry-project-document-ingestion

Run insight generation for a grid (enqueue document jobs and mark cells generating). This is NOT calculate-insight-grid, which only aggregates existing values. The Projects App calls this after the user acts in one insight grid. If the user wants to edit columns, rows, cells, or run insights in the UI, you call open-project-insight-grid. CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`.

run-insight-grid

You search only CJEU (Court of Justice of the European Union) judgments in the EU corpus. You do not use search-case-law or search-specific-court-case-law for CJEU judgments — those tools target national domestic courts only. You do not use this tool for EU acts text — you use search-eu-law. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-ecj-judgments

You search only ECHR (European Court of Human Rights) judgments. You use this tool for Convention rights, Strasbourg standards, or HUDOC-style questions. You do not use this tool for domestic national courts only. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-echr-judgments

You search only EU law / EurLex acts (regulations, directives, decisions, treaties) in the EU corpus. You do not use this tool for CJEU case law — you use search-ecj-judgments. You do not use this tool for national laws — you use search-laws / search-legal-acts. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-eu-law

You search national domestic court case law. You use this tool for case law on a topic, or to verify relevance, validity, and current status of specific cases (for example overturned, modified, or superseded by new practice). If the user names a court or court tier that has a court code in the active legal jurisdiction, you use `search-specific-court-case-law` for it. For all other domestic court practice, you use this tool. If the query is about CJEU (Court of Justice of the EU) judgments, you use search-ecj-judgments. If the query is about ECHR judgments, you use search-echr-judgments. Search in three time periods if needed: 1. First call with dateFrom set to 3 years ago. 2. If no results, call again with dateFrom 6 years ago and dateTo 3 years ago. 3. If still no results, call one last time with dateTo 6 years ago (no dateFrom). You try all three time periods before you stop. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-case-law

You find a court judgment document whose title equals the case number, then you search inside it. You use this tool to verify relevance or find specific quotes within a court case mentioned in a document or prompt. If the user names the court, you pass it as `courtName`, because several common courts use the same case number. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-chunks-from-court-case-by-number

Vector search restricted to a single legal document. Pass documentId as the opaque token from a prior search result (MCP link token or thread token), or a /mcp/document/<token> URL — not a raw UUID invented by the model, and not a /mcp/chunk/<token> path token. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-chunks-from-document-by-id

Resolve a document by title (embedding match), then run a vector search restricted to that document. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-chunks-from-document-by-title

You search national laws (documents with type 'law'). You use this tool when you need information in laws. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-laws

You search national legal acts from government institutions, ministries, and other state bodies. You use this tool when you need laws and regulations from the active legal jurisdiction. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-legal-acts

Broad search across national legal acts, EU law, case law, and municipal regulations when the document type is unclear or the query spans several source kinds. Prefer dedicated EU/ECJ/ECHR tools when those corpora are the explicit target. You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-legal-sources

Search municipality institution documents (vector DB). You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-municipalities-documents

Search documents published by a specific organization (resolve organization by name, then vector search). You write the vector search query in the language of the active legal jurisdiction. You do not write it in the user's UI language or message language. If the user wrote in another language, you translate the legal hypothesis before you search. Expand a code abbreviation to the official act name. Do not invent a statute article title. Do not remap a numbered article to a different topic. If the query names a code and an article number, you search that code and that number first. Do not add a guessed title. To batch several alternative queries in one call, you keep one in vectorSearchQuery and you put the rest in additionalVectorSearchQueries. They are searched in parallel, round-robin interleaved, and deduped. Each retrieved snippet includes matchedQueries. Each distinct query adds up to 6 snippets (maximum 18 total). CITING SOURCES: In your answer, ALWAYS link every source you rely on whenever a URL is available. When a result item includes a `sourceLink` field, cite it as a markdown link using the exact Vasara URL. For organization website results, link the exact external URL returned. Never present a legal source as plain text when a link can be included, and never invent or alter URLs. ## Vasara source links - Legal-corpus `sourceLink` values are authenticated Vasara URLs (`/mcp/chunk/<token>` or `/mcp/document/<token>`). You copy the URL verbatim into citation markdown links. You never invent or alter the URL. - Project, project-document, and insight-grid `sourceLink` values are authenticated Vasara URLs (`/mcp/project/<token>`, `/mcp/project-document/<token>`, `/mcp/insight-grid/<token>`). You copy them verbatim. They open the existing project UI after login. - When `citationMarkdown` is present on a project, project-document, or insight-grid item, you paste it. - You do not invent links. You do not use `/?chunk=...` paths (those are for the in-app assistant only). - You do not construct `/mcp/chunk/...` from `citationToken`. You always use the full `sourceLink` from the tool result. - Organization website search results still use the exact external `https://` URLs returned by the tool. You do not rewrite those into Vasara links. - For follow-up search inside a document (`search-chunks-from-document-by-id`), you pass the opaque `documentId` / `resolvedDocumentId` field from tool results — not a chunk path token from `/mcp/chunk/...`. A `/mcp/document/<token>` URL is acceptable when that is what the prior tool returned as `sourceLink`. Before your first legal search in a conversation, you call `get-legal-research-instructions` once. It returns the active legal jurisdiction, its query language, its courts and tools, and the full research protocol for this user.

search-organization-documents